Hi Abdelrahman,
> Is there a better way you can suggest using SPEC2017 in FS mode to > evaluate my system on gem5? > No, there's not a better way, as far as I know :). > My understanding is that the FS simulation is significantly slow. Any > guidelines on how to make it as fast as possible? You can use sampling methodologies, but there are tradeoffs there. You can run different simulation configurations in parallel as well. Generally, there's nothing easy to do to improve simulation speed. Jason On Thu, Mar 10, 2022 at 7:17 AM Abdelrahman S. Hussein < abdelrahman.sob...@gmail.com> wrote: > Thank you so much. > > There were some questions that you missed at the end of the email. I list > them below again for convenience: > > > - Is there a better way you can suggest using SPEC2017 in FS mode to > evaluate my system on gem5? > - My understanding is that the FS simulation is significantly slow. > Any guidelines on how to make it as fast as possible? > > Above all, thanks a lot for taking time helping us. Much appreciated! > > > > > On Thu, Mar 10, 2022 at 7:01 AM Jason Lowe-Power <ja...@lowepower.com> > wrote: > >> Hi Abdelrahman, >> >> From the get error message, it says: >> "This error may be caused by a too restrictive setting >> in the file '/proc/sys/kernel/perf_event_paranoid' >> The default value was changed to 2 in kernel 4.6 >> A value greater than 1 prevents gem5 from making >> the syscall to perf_event_open" >> >> If you put `1` or `0` in /proc/sys/kernel/perf_event_paranoid it should >> work. >> >> Cheers, >> Jason >> >> On Wed, Mar 9, 2022 at 9:27 PM Abdelrahman S. Hussein < >> abdelrahman.sob...@gmail.com> wrote: >> >>> Hello, >>> >>> I am trying to run gem5 in FS mode using the following command (I am >>> using gem5 v21.2): >>> >>> gem5/build/X86/gem5.opt >>> gem5/configs/example/gem5_library/x86-spec-cpu2017-benchmarks.py --image >>> spec-2017/disk-image/spec-2017/spec-2017-image/spec-2017 --benchmark >>> 503.bwaves_r --size test >>> >>> However, I got the following error (There are details after the error, >>> please continue reading after you check the error): >>> >>> build/X86/sim/simulate.cc:194: info: Entering event queue @ 0. Starting >>> simulation... >>> build/X86/cpu/kvm/perfevent.cc:176: panic: PerfKvmCounter::attach >>> recieved error EACCESS >>> This error may be caused by a too restrictive setting >>> in the file '/proc/sys/kernel/perf_event_paranoid' >>> The default value was changed to 2 in kernel 4.6 >>> A value greater than 1 prevents gem5 from making >>> the syscall to perf_event_open >>> Memory Usage: 3817816 KBytes >>> build/X86/cpu/kvm/perfevent.ccProgram aborted at tick 0 >>> :176: panic: PerfKvmCounter::attach recieved error EACCESS >>> This error may be caused by a too restrictive setting >>> in the file '/proc/sys/kernel/perf_event_paranoid' >>> The default value was changed to 2 in kernel 4.6 >>> A value greater than 1 prevents gem5 from making >>> the syscall to perf_event_open >>> Memory Usage: 3817816 KBytes >>> ./spec_fs_run.sh: line 16: 1648681 Aborted >>> >>> I have done some checking: >>> >>> - Ran kvm-ok to make sure that KVM is working: >>> $kvm-ok >>> INFO: /dev/kvm exists >>> KVM acceleration can be used >>> >>> - Also, I checked the value >>> inside /proc/sys/kernel/perf_event_paranoid >>> $cat /proc/sys/kernel/perf_event_paranoid >>> 3 >>> >>> - The SPEC17 is created by following the instructions in: >>> https://gem5art.readthedocs.io/en/v1.0.0/tutorials/spec2017-tutorial.html >>> >>> >>> Questions: >>> >>> - How to solve this error? >>> >>> - Is it safe to modify the /proc/sys/kernel/perf_event_paranoid of >>> the linux running on my host machine to 1, in case this solves the >>> problem? >>> >>> - Is there a better way you can suggest using SPEC2017 in FS mode to >>> evaluate my system on gem5? >>> >>> >>> >>> Thanks. >>> >>> -- >>> >>> *Best,Abdelrahman Hussein* >>> >> -- > > -- > > *Best,Abdelrahman Hussein* >
_______________________________________________ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
